MIT OpenCourseWareSession 6: Ponzi
MIT OpenCourseWare
Modeling challenges 0:00
The discussion opens with the main problems people ran into while building a Ponzi model. Timing was hard, especially when promised returns depended on when each person invested and what they expected back. Trust was another issue. If investors could not see inside the business, the only clear signal might come when payments stopped. Some people also found the individual-level version too expensive to run or hard to make fit other cases like Social Security.
Choosing aggregation 5:00
The next topic is the choice between aggregate models and agent-based models. Detailed models can feel natural because people, animals, and houses are concrete. That can pull you toward too much detail before you know it is the right level. Detailed graphics can also bias audiences and modelers toward agent-based work. The better choice depends on the problem, the cost of many simulations, and whether you need to see distributions, networks, local diffusion, or highly tailored policy effects.
Investor groups 19:33
The models sort people into clear groups: mature investors, new investors, former investors, potential investors, and people who are not willing to invest. The same split can be done with the money side too, as cash on hand and liability, or money owed at maturity. That keeps the accounting simple and makes the flow of people and funds easier to track.
Trust and withdrawals 29:30
Trust can be its own stock, or it can be folded into the way people move from skeptical to willing. If you model trust directly, you need an initial trust level and a way for it to change over time. The cash-and-liability setup is then the clearest place to look first, because it shows how new liability is issued and how withdrawals depend on confidence.
Redundant confidence terms 40:31
Keep one confidence construct, not two. A separate sensitivity term and then another parameter on top of it only create redundant parameters. That makes later estimation and interpretation harder. The cleaner choice is to put the sensitivity inside the confidence equation and then use that single concept directly in the model.
Liabilities and repayment 42:00
Liability means how much Ponzi owes people, not a normal risky investment. The scheme promised fixed returns, like 45 days for 50 percent. The model is useful, but it misses maturity. In the story, people had to come back to collect, then decide whether to reinvest. That delay can create bottlenecks and a sharp drop when cash runs short.
Asymmetric trust delay 59:30
Rahmandad explains that trust in a Ponzi scheme can change fast when money stops going out or journalists start asking questions. The spread of that news is delayed, but the delay should be asymmetric. People speak slowly when things look fine, yet they warn others quickly when payments fail. He says that can be modeled with a short delay when confidence falls and a longer one when it rises.
Albans model choices 1:03:01
He then reviews Alban's model. It uses a Bass diffusion setup, where adoption depends on claimed interest and confidence built from the fulfillment ratio. He likes the asymmetric adjustment, but questions some details. Debt should not keep growing forever after a term ends, and people returning to invest should also be limited by whether they were actually paid. A quick simulation shows confidence rising slowly and then crashing hard.
Reinvesting and payouts 1:21:33
The scheme keeps cash falling until it reaches zero, then investors wait to be paid. That waiting itself raises the desired cash out, because people want settlement. A cleaner model would also show the investment’s real return and compare it with an outside option, so the adoption fraction is not only built into sensitivity settings.
Model fixes and accounting 1:24:00
The model review then turns to one student’s setup. The payout path is tied to cash on hand, while the loss from unpaid obligations is tracked as scam damages. Reinvesting is better handled as a two-flow process, or as an auxiliary variable, because it should not confuse the investor stock. Liabilities matter here mostly for accounting and for showing when promised payments are not made.
Reinvestment and cash flow 1:42:07
The discussion returns to the collapse in the model and ties it to the fraction of obligations met falling over time. The reinvestment share is small in the long run, even when some people keep putting money back in after they are burned. That is treated as useful, since the scheme has already crashed, and Ponzi stays hand to mouth.
Investment timing and trust 1:46:00
The model can handle people who reinvest only part of what they get by treating that as a mix of full reinvestment and no reinvestment. The harder issue is changing average investment size over time. People start small, then put in more as trust grows. A cleaner way is to let trust drive the average amount, while keeping separate groups for small investors and large ones.
Ponzi collapse 2:02:33
Ponzi can keep growing while new investors come in and reinvest their payouts. Growth turns to slowdown when population growth becomes the limit and the cash pool no longer rises fast enough. Once one promised payment is missed, the collapse switch flips and the whole scheme stops, because nobody accepts investor equity anymore.
Market growth setup 2:07:07
The next model is a monopolistic firm selling a durable product in a limited market. Orders create backlog, delivery delay hurts sales, and the firm expands factory capacity when delays rise. You should separate work in progress from finished factories, include depreciation, and think about how sales force spending and delay together can produce ups and downs.
Two customer groups 2:23:00
He suggests modeling luxury brands with two customer groups and a simple stock-and-flow structure. Price, quality, and exclusivity shape whether each group adopts the product. High-end buyers want exclusivity, while lower-willingness buyers can be drawn in by price and wider use. The key question is not total profit, but whether a brand can thrive in its chosen segment and still keep a foothold elsewhere under the right conditions.
Books as status 2:27:30
For books, he pushes back on a pure format story and says the puzzle is more about status. Sales can rise because books, and especially BookTok books, have become a social signal, while reading rates fall for a separate reason. He suggests the model may need to capture broad adult decline and niche communities that read more, since the survey covers the general population rather than those groups.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
